自适应半监督邻域聚类算法

Adaptive semi-supervised neighborhood clustering algorithm

为了充分利用监督信息指导聚类过程,提出自适应半监督邻域聚类算法(adaptive semi-supervised neighborhood clustering algorithm, SSCAN)。引入监督矩阵与距离度量结合,构造合理的相似矩阵;充分利用监督信息,通过标签信息矩阵与流形正则项结合调整模型,改善聚类效果。在多种数据集进行试验,并与其他聚类算法作对比,结果表明,SSCAN可以充分利用监督信息,提高聚类的准确率。

In order to make full use of the supervision information to guide the clustering process, an adaptive semi-supervised neighborhood clustering algorithm(SSCAN) was proposed. The combination of supervision matrix and distance measurement was introduced to construct a reasonable similarity matrix;The supervision information was fully utilized to adjust the model through the combination of the label information matrix and the manifold regular term to improve the clustering effect. Through experiments on various data sets and comparison with other clustering algorithms, the results showed that the SSCAN algorithm could make full use of the supervision information and improve the accuracy of clustering.